Hi All,

I am working with a boutique hardware development company to development a custom open source linux-based router for production.  Our goal is to provide a powerful, flexible and compatible router for a competitive price.

We are convinced that the market needs a solid hardware platform, at a fair price, for the open-source community.  Thanks to communities such as openwrt.org we have a forum to discuss this.

Chip Manufacturer:

Q1.  What do you think, RealTek, Broadcom, other?

We know this is a hard question to answer, since we haven't narrowed it down to a certain SoC.  We have been looking at the BCM5354.  This seemed to be the most integrated solution and would allow us to keep cost down.  But, with a more integrated solution it looks like there are trade offs with performance. 

Q2.  Balancing Cost vs Performance, would you prefer a more integrated SoC that contained all functional components with lower cost and lower performance or a higher performance router with separate Switch, Radio and CPU switch at greater cost?

Q3.  We are also looking at providing a USB port.  We know that many people in the community like this option (external HDD, Flash Drive, etc).  However, not everyone needs this function.  Is adding the USB interface desired?

Q4.  WiFi, do we need to support a newer WiFi technology such as 802.11N?  Is it worth the uptick in price? Or simply use 802.11G?

Q5.  802.3af PoE - Given that this would add relatively significant cost, is this worth it to the community?

We are looking at this as just a good start on getting feedback.  Please help us by letting us know what you think and what you need.

Thanks For Your Help!

Definitely a must!

When N is out, B & G will be phased out.

I won't consider to purchase my next router without this support. It will make life easier.

Hi Mazilo,

All good feedback.  I will get some more information on PoE costs to give you and the community an idea of what the actual cost will be so we can evaluate this option.  It sounds like many people will want this feature.

Thanks again for the feedback,
Phiz

For a low cost SoC with good all-around performance I'd suggest Atheros.  If you want absolutely abysmal wifi performance Realtek can sell you some.  Broadcom's driver situation sucks and I don't foresee that changing anytime soon.

You can't make everybody happy with one board.  I'd personally like a gig-e switch but I can't see hanging a gig-e switch on a machine that can only route a few mbit before choking down.

I personally wouldn't use it.  USB-based wifi adapters generally lack performance and drivers.   USB based cameras are overpriced and generally have awful output quality (and a lack of Linux drivers)...  I can't see the point of trying to pretend its a file server if you're suggesting the use of a typical 200-300 mhz MIPS SoC.  Theres just not enough performance.

how about an 'internal' express slot (whatever they call the laptop version of PCI-E)?  I'm kind of amazed that people are still implementing oldschool PCI on SBCs.  PCI-E requires like 1/10th the wiring as PCI.  This would also take the FCC responsibility out of your hands (faster time to market).

passive PoE is fairly affordable to implement.  Theres no reason for active poe unless you're suggesting gig-e interfaces.

My only suggestion is to bring the GPIOs and serial to proper header(s).  This will let the buyer add things (SD Cards, LCDs, etc) without a fight.  If the SoC has USB support, I'd suggest bringing it to a cheap-to-implement pin header (standard mobo layout) so the user can choose if he wants to spend the money/waste the space on the USB header.

If it were up to me, I'd start with dime-a-dozen bulletproof hardware and increase capabilities from there.  Especially coming into the equation essentially as an unknown, first-time customers are far more likely to spend their marginal cash if the cost of both entry and resultant failure is low.  This is why the *wrt distros were so successful, is because the base Linksys (and other) devices were so initially cheap, no one thought twice about bricking a $50 piece of hardware.

I view these as one question, and refer to my above note.  I personally would prefer something with integrated ADSL and a switch, but that's just me.  It is my opinion that all the file server, media system, and similar extensions have gotten too far away from the original premise: a purpose-built piece of router hardware with custom firmware.  I don't dislike the additions, but disagree with how much focus they've taken off of the basic functionality.

Attractive, particularly if a chipset w/integrated ADSL (or Cable for those that swing that way) is not a viable option.

Coming from a Wi-Fi 'professional', N is nice but G will get you much more bang for your buck for the coming months.  If I need the speeds N professes to provide, I'll wire my clients.

Again, a nice-to-have but absolutely unnecessary for core operation.  This would fill the needs of a very small but unusually vocal crowd of people trying to shoehorn consumer equipment into professional capacities (and an even smaller group of prosumers that have convinced themselves running 48VDC @ 500mA over amateur, stranded 28ga cabling is somehow a good idea).

You'll have to balance between the first-adopter crowd that wants a quad-core Opteron in a Gumstix form factor and the real needs of the remaining 95% of the population.  If you pursued a core SoC that has a PCI bus (does a good one exist?), it would be easier to build more fully-featured models later while maintaining the cheap, bulletproof core.

I am just curious. Will a 200-300MHz SoC have enough juice to drive this PCI-E for a max. performance?

It is important that the used SoC are available more than a few months.

Enough power for the packet filter on maximum network throughput and applications. My provider for example will start with 100MBit symmetric fiber optic connections until summer. I would prefer a design with a more integrated SoC and a (additional?) memory slot. Minimal flash size for boot loader only and internal USB for the drive?

USB ports are important. Some users use USB DSL modems.

Nice to have but not really important.

Additional whises:
- hardware crypto engine for IPsec and WLAN encryption
- (second?) miniPCI to add a ISDN card?
- optional boot loader with PXE support

This is great input J4k3, aoz.syn, mazilo and norbert. 

I want to discuss this in more detail with the other business partners, but I tend to agree with all of you.  From my personal point of view, I want all of the options, but I think that aoz.syn brings up some great points about cost, time to market, etc. 

c brings up many good points.  Regarding PoE, as long as we stay within the 12.9W power budget, we should be able to have a separate PoE splitter for those who wish to use PoE.  I think these typically run around $15 on the www.   Would this work good enough for those that need PoE?  The good thing about this is that not everyone has to pay the extra charge for PoE.

Norbert, for the users needing USB for DSL, do you think we need to incorporate multiple USB ports or do you think it will be fine with an external USB hub?  Also, I personally don't have a lot of experience with IPsec crypto resource rquirements.  I am assuming that performing these functions in software will severely limit us??

Mazi, I will look up more about resource requipments of PCI-E and whether a sub-300MHZ cpu core can hang.

J4k3,  the broadcom drivers issues.  Are they just not as forthcoming or is it an overall compatibility issue?  Also, you had suggested Atheros for a low-cost good all-around SoC.  What chipset would you recommend?

I know I haven't quite done your post justice quite yet.  I will continue to work with the team and update you guys.  Thanks again and please keep the feedback coming!

Thanks!
Phiz

I've just seen too many little hardware shops with all the right ideas but no customers strangle themselves trying to bite off the super-powerful, all-in-one widget on their first try.  IMHO, it is better to have a steady income stream from an established, appreciated product and develop from that than to waste all your capital doing the big one.

Absolutely - build in the circuitry to use it if you can do so economically, but leave the injector as a value-added option.

Put in as many as you can within the price and power envelope you settle on, and leave it at that.  If using a USB connection, I personally would rather have one 2.0 port and have to provide a hub than have two 1.1 ports.  Another value-added opportunity: design a USB hub that will physically snap to the core device.

Yes & no.  Most current SoCs seem to have sufficient speed to satisfy the majority the SOHO market without acceleration, but there are definite edge cases.  If you do something based on VIA or Intel IXP, you'll have built-in crypto; otherwise, just add a peripheral bus down the road (eg mini-pci) so people that really need the acceleration can add in something like the Soekris VPN1411.

Lack of openness.  A lot of the OSS support for Broadcom anything these days is reverse-engineered.

I would second J4k3's note on GPIO, USB, and serial headers - terminate them sensibly (0.1" spacing) and in a well-documented location.  Even if they're just through-hole endpoints, pins and solder are cheap!

I recently bought an Asus WL-500-GP, for these reasons:

- it has 8MB flash (as opposed to only 4 or even less)
- it has 32MB ram (as opposed to only 16 or even less)
- it's supported by openwrt

Several other people I talked to in #openwrt bought the same device for this reason, as well. It's a very popular router, but apparently it was too expensive in production, at least the new V2 version of this router seems to be dumbed down a lot.

Regarding your other requirements, please note that I'm not involved in development at all, I'm just a regular user:

Q1: don't know. it does not matter much to the end user; the requirement here is: it has to work fine. I understand that the driver situation is not the best for my ASUS router, but it does work (with 2.4 kernel). So minor problems are not necessarily a show stopper as long as it can be supported, at all.

Q2: I don't mind a single chip / low cost solution, as long as it works. I never opened my router and I don't intend to do so in the feature. Although I could replace my wifi minipci card which is apparently in my router, I don't intend to do so, ever.

Q3: my router has USB and I never used it. if it can actually be put to some use (download on usb stick, server for usb printers and scanners), that'd be nice to have. But it's really way down on my personal wishlist.

Q4: newer wifi technology: none of my other wifi devices support it either, so this would be completely useless

Q5: no need

Regarding headers:
I'm not interested in boards. If you have to buy a board and a case separately, it's just too expensive. I prefer a box that comes ready to use.

If you make a router, that is supported by openwrt and other free firmwares, and that is not tight on resources, and does not have a completely unreasonable high price compared to other products, you should have something that sells.

Two external ports are enough. I use currently on for a SpeedTouch modem, the firmware can be found in the package trunk.

You should have a look to www.danm.de. We use at the moment ASUS WL-500gP to establish StrongSwan IPsec VPNs from and to customers. But without crypto hardware acceleration with a maximum of two connections. With crypto hardware acceleration the router can replace proprietary gateway and VPN appliances like a SonicWall. I don't trust these appliances, no documentation abount content scanning, no technology updates like IKEv2 and so on.

The ASUS WL-500gP is a good template. Broadcom cores are not perfect but usable.

At the risk of being contentious for a cause that is not mine to decide, this is precisely what I was talking about earlier - people shoehorning consumer equipment into commercial capacities and being disappointed when it doesn't perform like commercial equipment.  There already exist very usable competitors to SonicWall et. al. in the WL-500gP's price range, such as the ALIX.  I believe it would be unwise to jump into the market right in the middle of the most heavily-contested price point, US$100-$120.

Crypto accelerators are EXPENSIVE - I've done a good deal of research into this, and short of the VIA and Intel processor-integrated engines, you're looking at a minimum of $70 per board (less for the chips in bulk, of course, probably $45-50), and that's for a mini-PCI Hifn 7955, not exactly the latest.

As an alternate discussion, norbert, if you really need VPN performance out of the WL-500gP and can live without the wireless functionality, gut the radio card and antenna and put a Soekris VPN1411 in their place.  The 2.6.25 kernel just got a native driver for them.

Nah, it'd have the same problems the PCI ones do now.  a PCI-E 1x channel isn't any faster than a regular PCI channel, just each PCI-E channel has its own bandwidth (that can be bonded)

If you really are into this and since OpenWRT also supports x86 platform, why not design your router based on an existing AMD 64- X2 board with PCI-E on a micro footprint board with standard RAM sockets, except run it @ a much lower frequency, i.e. 500MHz, to avoid using fan? This way, this way the router can easily run encryptions. You can sell the router equipped with a 32MB DRAM. If end users want to add more RAM, they can just buy the new RAM from local stores to replace the existing one easily.

(Last edited by mazilo on 3 May 2008, 00:42)

Hi All,

Again thanks for the comments.  Each day there is more and more useful information that you are providing.  I am going to have a full review with the rest of the team here on Tuesday and will reply with the outcome of that meeting.  We will probably have to make some general decisions (single SoC vs multi-chip and which chip manufacturer, WiFi standard, etc)  within the next 1 to 2 weeks.  How much Flash, RAM, etc can come later to a certain extent.

This is really going to happen, so if you don't provide input, don't complain later!

Have a good weekend,
Phiz

This will only be good if one plans to design a router with more than one PCI-E channel. If you ever have a need for more than one PCI-E channel, why not just build your own router with an AMD-64 X2 using a micro-footprint board?

Can this be done more efficiently with a TI AR7 chipset?

I know that crypto cards are very expensive. We don't found a good solution for our problem. But your Soekris hint is very good. I order one miniPCI card on monday for testing. You should add a extra miniPCI slot.

(Last edited by norbert on 4 May 2008, 10:51)

Having a USB or RS232 port is essential for me as I am using wifi routers as serial servers.  Sticking a USB port on a wireless router creates so many possibilities I would never buy one without now as it would be crippled to me!

From the feedback here, we will definitely plan on having a USB port (how many and 1.1 vs 2.0 is open for discussion).  If we just have one, we will do everything we can to provide a 2.0 compliant port.  If we have more than one, we may look at 1.1, 2.0 is proffered though!

Also, we are looking at using the MAXIM RS3232 IC onboard to convert the TTL over to a DB9 Female.  Sound ok?

Thanks,
Phiz

(Last edited by phizaleo on 6 May 2008, 04:42)

I like the BCM5354. Integrated solution does not necessarily mean poor performance The 5354 has separate ethernet for lan and wan which adds up to improved performance. I also like the integrated usb 2.0 support. This all argues for lower power consumption for applications like an autonomous robot. 240 MHz to boot.

(Last edited by tetelestai on 6 May 2008, 19:53)

I think you mean the MAX3232.  This chip will actually support two RS-232 ports, one would obviously be great for the console, but I would at minimum provide a header for the user to add their own DB9 later should they wish to use another serial port for some purpose.

With regards to the USB port, if you're providing one, providing a second should cost little more, it's just a matter of taking up the space for the second connector.  I'd definitely consider making these 2.0 and not 1.1.  Most everything that I've seen users wanting to add (Ethernet, webcam, hard drive) will benefit from the speed.

One additional thing not mentioned here would be a real time clock, another nice to have goodie that makes life nice.

just throwing few thoughts....

I think by going for a "one size fits all" will be a real challenge.... I doubt you can or would make anyone happy at the end...
Maybe it would be clever to "split the path" and make more (possible) users/customers happy...

First of all, I think modularity and "openess" is a must. As already suggested, at least make things accessible via pin headers... making the solution modular would allow the users to actually choose what they need and pay accordingly.
So, on one side you could have a "small & compact" solution targeted at low price with "not so much horse power" etc.. and a "pro" version with more powerful cpus, maybe with some accelerators, more expandability, more ram...

Talking about the "little one" I think a SoC solution is the right way to go something integrated as much as possible with "goodies" not available elsewhere:
- SoC solution with integrated BG radio, 1 (+1) LAN, serial, usb, ~8MB Flash, ~32MB RAM - but make thinks modular - give out just the "bare minimum" with upgrade options - a nice feature would be a miniPCI/PCIe slot, to be able to upgrade the radio or add other "functionality", include serial, usb, second lan.. maybe in form of "upgradeable modules" or just leave the pin headers or sell additional "DIY kits" for upgrades

The "big one" should have a powerfull cpu (IXP4xx, x86...) so various ipsec or other compute-intensive tasks will feel comfortable, add few *real* ethernet ports (maybe some GigE also) more ram (in a sodim?) and more flash, maybe an CF slot... at least 3-4 expansion ports (miniPCI/PCIe)

This way you could really make people happy. If you could offer all that for a "really reasonable" price...

The PoE, maybe just the passive one, is very important as I see it. And you can simply make it modular again so users have a choice in terms of price/performance...

Anyway, best wishes for this project! I really can't wait to see what will get out of it... if possible, I'm just signing for samples right now.

Best regards,
M.Culibrk

I strongly agree with these specs and they are mostly readily available on x86 platform as a computer mobo, except you will need to shrink it down to a small size (preferably smaller than the existing micro board). The benefits on using this approach is the design is freely available and it comes with HD controller + Video port.

As with firmware, you can have OpenWRT and/or any open-source firmware communities developed it to cut down the costs. This means, your hardware specs need to conform to at least OpenWRT specs.

Build it like the wrt54gl but with a stronger transmitter and only 2 lan ports (one for DMZ and one for LAN to another switch I already have).

But keep the end user price at $40 (same as open-mesh.com is offering).